Rapid progress in large language model reasoning has driven the 90.5% market-implied odds for an AI system reaching ≥90% on FrontierMath before 2027. Top models such as Claude Fable 5 and GPT-5.6 variants already score 87–89% on updated tiers as of mid-2026, up sharply from under 2% at launch in late 2024 and roughly 40% earlier in the year. This trajectory stems from iterative scaling, improved chain-of-thought techniques, and specialized math training that narrow the gap on research-level problems. Trader consensus views continued releases from leading labs as likely to close the remaining distance within months. Still, realistic headwinds include potential benchmark revisions that raise difficulty, unexpected capability plateaus, or slower-than-expected gains on the hardest Tier 4 problems.